DeepSearch, Minimum code to implement Deep deep using deepseek LLM (最小化代码实现deepsearch)

DeepSearch is a Python package that combines data science utilities with powerful search and LLM capabilities. It allows you to create ReAct agents that can search for information and reason through complex questions.

Features

• ReAct Agent: Implements the Reasoning and Acting (ReAct) pattern for LLM agents
• LLM Integration: Connect to various LLM providers including DeepSeek and local models
• Search Capabilities: Integrated search functionality using the Serper API (Google search)

Installation

# Clone the repository
git https://github.com/ranchlai/DeepSearch.git
cd deepsearch

# Install the package and its dependencies
pip install -e .

Usage

Basic Example

from ds.search import react_agent

# Run a ReAct agent with a question
question = "小神童3号可以在平安好车主app上购买吗?"
answer = react_agent(question, max_steps=5)
print(answer)

Flask Web Service

You can run DeepSearch as a web service using Flask:

# Run the Flask app
python app.py

This will start a web service at http://localhost:800/search that accepts GET requests with a query parameter 'q'.

Example:

http://localhost:800/search?q=小神童3号可以在平安好车主app上购买吗?

The response is returned in JSON format:

{
  "query": "小神童3号可以在平安好车主app上购买吗?",
  "result": "根据搜索结果，没有发现小神童3号产品可以在平安好车主app上购买的直接信息..."
}

Environment Variables

The following environment variables need to be set:

• DEEPSEEK_API_KEY: Your DeepSeek API key
• JINA_AI_API_KEY: Your Jina AI API key for search, you can get it from here
• SERPER_API_KEY[Optional]: Another option is to use Serper API key for search, you can get it from here

License

Apache License 2.0